01
67
Cites
Responsible AI · AI Governance for Development
Nasim Mahmud
Nayan
Currently
Deputy Manager, AI & ML · BRAC
Making AI accountable where the stakes are highest — healthcare, education, and development at national scale.
12
Publications
174
Citations
6
h-index
4+
Years in AI / ML
10k+
Students Reached
01
production engineering
About
Research depth meetsproduction engineering
I work on the standards an AI system must meet before it reaches frontline work at population scale — where fairness, evidence, and accountability stop being research questions and become deployment constraints.
My published work reduced demographic bias by 30.8% in clinical models with no loss of performance, and cut cross-modal disparity from 7.6% to 4.5% in multimodal diagnosis. At BRAC I apply the same standard to deployment — shaping policy on responsible AI use, data protection for AI products, and the evidence a system must produce before it reaches people. The research and the governance work are the same problem approached from two ends.
Growing up in rural Bangladesh, where healthcare access was limited, I saw at first hand the need to democratise medical care and education. That experience led me to found OgroPath, an AI platform making medical entrance-exam preparation accessible to students nationwide.
Research Focus
AResponsible AI Governance for Low-Resource Settings
Policy, data protection, and evaluation standards for AI deployed at population scale in development contexts.
BFairness as a Deployment Constraint
Bias detection and mitigation across demographic, annotation, and amplification axes — treated as a release gate, not a research finding.
CTrust & Interpretability in Large Language Models
How trust signals transfer, and what makes a model's reasoning auditable rather than merely explainable.
DCitation-Grounded Retrieval & Decision Support
Source attribution and fact-checking for clinical and enterprise question answering.
EMedical AI & Cyber-Physical Systems
Multimodal diagnostics and privacy-preserving pipelines for maternal health and remote monitoring.
02
h-index 6 · i10 5
Selected Publications
12 papers · 174 citationsh-index 6 · i10 5
02
23
Cites
03
Date Fruit Classification with Machine Learning and Explainable Artificial Intelligence
19
Cites
04
An IoT-Based Real-Time Environmental Monitoring System for Developing Areas
18
Cites
05
9
Cites
06
7
Cites
07
Investigation of Air Effluence Using IoT and Machine Learning
6
Cites
08
Air Pollution Monitoring Using IoT and Machine Learning in the Perspective of Bangladesh
5
Cites
09
Recent Advancements of Computer Vision in Healthcare: A Systematic Review
4
Cites
10
Enhancing the Security of Pregnancy Health Data Transmission through Homomorphic Encryption: An Advanced Model
1
Cites
11
Published
12
Designing an AI-Driven Logistics Planning Artifact for Constraint-Aware Vehicle Routing
2026
03
to founder-led products
Industry & Entrepreneurship
Enterprise AI strategyto founder-led products
BRAC — Central Data Team
Deputy Manager, AI & Machine Learning
May 2026 – Present
- Owning the AI and machine learning roadmap and exploring AI opportunities across BRAC's programmes and enterprises, so that frontline colleagues get support in their daily work where the need is real.
- Turning BRAC's existing data into a base for quick, evidence-led decisions, using AI and NLP to connect it to BRAC's five-year strategic goals as part of a targeting dashboard for leadership.
- Contributing to AI governance across BRAC by shaping policy on responsible AI use, data protection for AI products, and the standards a system must meet before it reaches frontline work at population scale.
- Overseeing AI products built by programme teams and specialist vendors across the model lifecycle, including a legal advisory chatbot for SELP and an agricultural advisory tool with the Social Innovation Lab, reviewing the architecture and problem fit and advising where the approach needs to change.
AI GovernanceResponsible AIData ProtectionModel LifecycleNLPCloudera
Solyntra Limited
AI Engineer
Dec 2025 – Apr 2026
- Led R&D and architecture for SaaS AI knowledge systems with citation-grounded RAG pipelines.
- Implemented metadata-aware ingestion, hybrid retrieval, and a 'No Source, No Result' grounding policy.
- Performed parameter-efficient LLM fine-tuning with LoRA (PEFT) and multi-agent architecture.
LLMsRAGLoRAHugging FaceFastAPIPostgreSQL
Programming Hero
Machine Learning Engineer (Remote)
Dec 2023 – Sep 2025
- Shipped Zenyora AI to the Microsoft Store — a wellness platform with real-time posture detection at 95% accuracy and sub-100ms inference, serving 100+ daily users.
- Architected a voice-enabled RAG real-estate sales assistant over 2,000+ listings at 95% query relevance using Whisper STT and ElevenLabs TTS.
PyTorchOpenCVLangChainPineconeGPT-4Docker
Primacy Infotech Ltd
AI Engineer
Jul 2023 – Nov 2023
- Led an AI tour planner for 6 UNESCO sites, cutting itinerary creation from 2 hours to 5 minutes.
- Built a clustering pipeline over 5,000+ visitor surveys, raising satisfaction scores 35%.
GPT-3.5FastAPIPostgreSQLDockerStreamlit
04
models · 3 research labs
Research
Trust signals in frozenmodels · 3 research labs
Current research programme
The Transferability of Trust Signals
When and why does a model’s trust signal carry from one task or language to another?
The transfer ladder
Paper 1 scope
Same language, new layout
Whether a trust signal established on one document layout still holds on a layout the model has never seen.
Rung 01 · document → document
Paper 1 scope
Across languages
Whether that same signal survives the move into a lower-resource language, where far less of the model’s training data lives.
Rung 02 · cross-language
Later work
Across paradigms
Document extraction to retrieval to agents. The hardest rung, and deliberately years out.
Rung 03 · cross-paradigm
Jul 2022 – Present
Medical Cyber-Physical Systems
Shanto-Mariam University of Creative Technology
Shanto-Mariam University of Creative Technology
Advisors — Prof. Mohammad Mobarak Hossain · Dr. Jasim Uddin
Research Assistant — Medical CPS
- Architected an ensemble framework for maternal-health risk stratification reaching 99% accuracy and cutting algorithmic bias 34% across 5,000+ records.
- Implemented homomorphic encryption on IoT devices for secure, compliant pregnancy-data transmission.
- Co-authored 3 papers in high-impact venues (combined 50+ citations).
Aug 2022 – May 2023
Healthcare AI & IoT Systems
Rising Research Lab
Rising Research Lab
Advisors — Md. Monirul Islam · Dr. Jia Uddin
Research Assistant — Healthcare AI & IoT
- Deployed a multi-disease prediction system (diabetes, Parkinson's, maternal health) at 92% average accuracy with a 5-location IoT air-quality network.
- Engineered a full pipeline from IoT capture to real-time inference.
- Built an AutoML pipeline that cut model-development time 60% while maintaining performance.
Jan 2023 – Dec 2024
Computer Vision in Medicine
EMPATHY Lab, Independent University Bangladesh
EMPATHY Lab, Independent University Bangladesh
Advisors — Dr. Ashraful Islam · Dr. Muhammad Usama Islam
Research Assistant — Computer Vision
- Conducted a systematic review of 125+ papers on healthcare computer vision, mapping key gaps and opportunities.
- Analysed CNNs and Vision Transformers for medical imaging and predictive analytics.
- Synthesised findings to steer lab direction in surgical assistance and remote monitoring.
05
production AI systems
Key Projects
Open-source toolkits &production AI systems
MarkFeed
Offline Document AIConverts scanned PDFs, digital PDFs, Word and spreadsheet files into clean Markdown — with extracted images, tables, per-page accuracy stats, and a verify view showing the original scan beside the converted text. Runs entirely offline on open-source OCR and layout detection, with first-class Bengali and English scanned-book support.
100%
Offline
0
LLM calls
BN + EN
Scripts
FastAPITesseractPaddleOCRimg2tablePyMuPDFDocker
M-TRUST
Open SourcePlug-in Python toolkit to detect and mitigate demographic, quality, annotation, and amplification bias in clinical AI — one-line wrapper API with docs and reproducible examples.
30.8%
Bias reduced
4-axis
Bias coverage
PythonPyTorchScikit-learnPandasMatplotlib
ExplainRAG-FC-AS
Clinical RAGClinical decision support via RAG & LLMs with thematic clustering and NLI fact-checking before generation — source-level attribution for traceable, transparent recommendations.
<10s
Query time
90%
Clinician pref.
0.85
Evidence conf.
FAISSK-MeansRoBERTa-MNLIGPT-4Streamlit
Trustworthy ECG Platform
Edge AIReal-time and offline ECG analysis with fairness-aware training and uncertainty estimation — supports live sensor streams and digitised paper ECGs with natural-language explanations.
>0.9
Fairness score
95%
Accuracy
<100ms
Edge inference
PyTorchResNetSciPyStreamlit
Multimodal Chest X-Ray Diagnosis
Bias AnalysisCombined CheXNet image features with BioBERT clinical text to classify 14 thoracic diseases — diagnosed 4.3× cross-modal bias amplification and added Grad-CAM explanations for clinician trust.
14
Diseases
7.6→4.5%
Disparity cut
TorchVisionBioBERTGrad-CAMDocker
RAG-Based Website Generator
Cost-Efficient AIAI website generator using RAG to assemble pre-built React components instead of generating code from scratch — with PostgreSQL + pgvector semantic search and tree-sitter AST parsing.
92%
Token reduction
94%
Cost savings
FastAPIpgvectorNext.jsGPT-4otree-sitter
★
Apr 2025 – Present
AI for Low-Resource Service Delivery
Founder & Lead AI EngineerApr 2025 – Present
OgroPath
AI-Powered Medical Exam Prep Platform
Founded to democratise medical education for students across Bangladesh, especially in underserved rural communities. OgroPath combines NLP-based question generation with adaptive learning to make medical entrance exam preparation accessible to all.
AI Question GeneratorNLP-powered from PDFs
21K+ QuestionsComprehensive bank
Adaptive LearningPersonalised study plans
National RankingReal-time system
21k+
Questions
10k+
Students
Active
Platform
LLMsNLPFastAPIPostgreSQLReactDocker
Visit OgroPath →
Social Impact Mission
"Growing up in rural Bangladesh, I witnessed the barriers to quality medical education. OgroPath removes these barriers by providing AI-powered, affordable exam preparation to aspiring medical students nationwide — making medical education accessible regardless of location or economic background."
06
& mentoring moments
Highlights
Competitions, recognition& mentoring moments
07
awards · teaching
Skills & Recognition
Technical depth ·awards · teaching
Capabilities
AAI Strategy & Governance
AI StrategyUse-Case PrioritisationResponsible AIAI GovernanceFairness & ExplainabilityVendor Assessment
BAI & Generative AI
LLMsRAGAI AgentsNLPComputer VisionPredictive ModellingRecommender Systems
CArchitecture & Tools
Enterprise AI ArchitectureData PipelinesModel LifecyclePyTorchTensorFlowScikit-learnHugging FaceLangChainClouderaMLflowDockerFastAPIVector DatabasesPostgreSQL / pgvector
DLeadership
Stakeholder ManagementCross-Functional DeliveryVendor CoordinationLeadership ReportingTechnical Mentoring
ELanguages
PythonC++C
Journal Peer Review
- International Journal of Human–Computer Interaction — Taylor & Francis (Q1), 2025–Present
- Informatics and Health — Elsevier, 2025–Present
Awards & Honours
- Employee of the Month — Primacy Infotech Ltd (Aug 2023)
- 2nd Place — Inter-university Programming Contest, UITS (2022)
- Fastest Problem Solver — UITS Victory Day Programming Contest (2021)
- 1st Place — Inter-university PowerPoint Presentation Competition, UITS (2020)
Teaching & Mentoring
- AI Trainer — EDGE Project: trained 50+ students in ML/AI (2023)
- Research Mentor — UITS Summer Research Program: 7 students, 1 paper accepted (2023 & 2024)
Teaching & Mentoring
50+
Students Trained
AI Trainer · EDGE Project
7
Research Students
UITS Summer Program
85%
Job Placement
EDGE Project Alumni
Education
B.Sc. in Computer Science & Engineering
University of Information Technology and Sciences (UITS)
Jan 2019 – Jun 2023
CGPA 3.62 / 4.00 — Ranked 6th in department
Thesis — A Multi-Disease Prediction Framework: Leveraging Machine Learning and Real-Time Applications for Improved Health Outcomes.
Standardised Tests
TOEFL98 / 120
IELTS6.5 / 9.0
GRE307 / 340Q153 · V154 · AWA 3.0
Let's build trustworthy AI
Let's work together.
Open to PhD positions in trustworthy AI, research collaboration on trust and interpretability, and conversations about AI governance for development.